New energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling

Abstract

The present invention provides a new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling, which aims to solve safety hazards such as thermal runaway caused by high-density vehicle gathering. The system includes a state acquisition module, a cluster identification module, a regional risk modeling module, a risk level determination module, a response control scheduling module and a thermal risk prediction module. The system collects operating status data such as vehicle position, battery temperature, and state of charge, uses a density-based spatial clustering algorithm to identify clusters, calculates the risk score value of the cluster, and implements current limiting and path guidance control according to the risk level. At the same time, the trained long-short-term memory neural network model predicts future regional risk change trends to achieve early warning and response. The system is suitable for safety management scenarios of new energy vehicles in high-density areas such as charging stations and underground garages.

Description

Technical Field

[0001] The present invention relates to the interdisciplinary technical field of intelligent traffic control and artificial intelligence algorithms, and in particular to a new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling. Background Art

[0002] With the widespread adoption of new energy vehicles (NEVs), centralized parking and charging systems are gradually forming in cities. This is particularly evident in underground garages, bus terminals, commercial complexes, and residential areas, where large numbers of NEVs are densely clustered. The dense parking and charging of multiple NEVs within a physical space can lead to the following safety issues: thermal runaway chain reactions; localized temperature rises that can trigger battery failure; and the accumulation of smoke, heat, and gases that can cause secondary hazards.

[0003] Existing technologies mainly implement thermal safety monitoring through the battery management system (BMS) of a single vehicle, or rely on traditional fire-fighting equipment for post-event disposal. Some platforms also attempt to control local temperature rise or charging power based on fixed thresholds. Related research mostly focuses on individual vehicle risk assessment or global charging load optimization.

[0004] However, these technologies face numerous shortcomings in practical application. For one thing, current methods generally lack the ability to spatially identify multi-vehicle clustering behavior, making it impossible to dynamically identify cluster patterns and their associated risks. Furthermore, risk assessment mechanisms often overlook the combined impact of vehicle status, temperature changes, and local environmental loads, making it difficult to manage potential risks in a hierarchical manner. Furthermore, existing response strategies are mostly based on static rules or manual scheduling, making it difficult to develop a real-time, efficient coordinated response mechanism or meet the requirements for early warning and proactive control in high-density scenarios.

[0005] To this end, it is urgent to propose a technical path for risk prevention and control in scenarios where new energy vehicles gather, so as to achieve dynamic management of the entire process from identification, assessment, judgment to response. Summary of the Invention

[0006] The present application provides a new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling to achieve real-time identification and early intervention of the risk of high-density gathering of new energy vehicles.

[0007] This application provides a new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling, including:

[0008] A status acquisition module is used to collect the operating status data of new energy vehicles in the target area, wherein the operating status data includes the vehicle's geographic coordinates, battery temperature data, state of charge data, and vehicle operating status tags;

[0009] A cluster identification module is used to identify clusters based on the operating status data using a density-based spatial clustering algorithm to obtain cluster characteristics of the clusters;

[0010] A regional risk modeling module is configured to calculate a risk score for each cluster based on the cluster characteristics provided by the cluster identification module; the risk score is obtained based on the vehicle density per unit area, the average temperature of the cluster, the proportion of high-temperature vehicles, the proportion of vehicles in abnormal conditions, and the charging load ratio of the target area;

[0011] A risk level determination module, configured to determine the risk level of the corresponding cluster according to the risk score value;

[0012] A response control scheduling module is used to execute flow control and path guidance measures according to the risk level and provide response results to the external platform;

[0013] The thermal risk prediction module is used to construct a time series feature dataset based on a sliding window according to the operating status data, and use a pre-trained long-short-term memory neural network model to predict the potential risk score trend of each target area within a preset time range in the future; the predicted potential risk score trend is output to the risk level determination module and the response control scheduling module to execute the current limiting or guidance strategy in advance.

[0014] The beneficial effects of this application mainly include: (1) It can identify the spatial aggregation behavior of new energy vehicles in real time, effectively extract high-risk clusters through density clustering algorithms, realize the technological leap from individual monitoring to group identification, and significantly improve the perception of vehicle aggregation risks in high-density scenarios. (2) By introducing a multi-factor risk modeling method, factors such as vehicle distribution density, temperature level, abnormal state ratio and regional charging load are integrated into risk scoring values, which improves the comprehensiveness and scientificity of risk assessment and avoids the misjudgment and omission problems caused by relying on a single indicator in existing technologies. (3) With the help of the risk level judgment results, dynamic current limiting and path guidance control are realized, so that the response measures have hierarchical linkage capabilities, effectively reducing the risk of thermal runaway in high-risk areas and improving the response efficiency and safety level of the system. (4) By introducing a thermal risk prediction module based on sliding windows and LSTM networks, the future risk evolution trend can be predicted, thereby intervening in potential aggregation risks in advance and enhancing the system's foresight and active prevention and control capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a schematic diagram of a new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0016] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific implementations disclosed below.

[0017] The first embodiment of this application provides a new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling. Figure 1 , which is a schematic diagram of the first embodiment of this application. Figure 1 The first embodiment of the present application provides a new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling, which is described in detail.

[0018] The new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling includes a state acquisition module 101, a cluster identification module 102, a regional risk modeling module 103, a risk level determination module 104, a response control scheduling module and a thermal risk prediction module 106.

[0019] The status collection module 101 is used to collect the operating status data of new energy vehicles in the target area. The operating status data includes the vehicle's geographic location coordinates, battery temperature data, charge status data and vehicle operating status label.

[0020] The status acquisition module 101 is used to continuously and in real time collect data from all new energy vehicles in the target area that are in operation, parked, or charging, in order to provide complete and accurate operating status data for use by subsequent processing modules. To achieve this goal, the status acquisition module 101 is deployed in designated monitoring areas in the city, such as underground parking lots, electric vehicle centralized charging areas, battery swap stations, bus terminals, and other places with high-density vehicle aggregation. The module can integrate a variety of acquisition methods, including fixed cameras, parking space detectors, charging pile interfaces, battery T-Boxes, on-board OBD terminals, LBS positioning devices, Bluetooth or ultra-wideband RTLS tag systems, etc.

[0021] During the specific data collection process, the status collection module 101 needs to complete information perception in the following dimensions:

[0022] First, the geographic coordinates of each new energy vehicle are collected. This coordinate information can be obtained using high-precision GPS, RTK, UWB, or base station fusion positioning methods, with units accurate to meters or sub-meters. It is used to describe the vehicle's static parking position or dynamic movement trajectory in two-dimensional or three-dimensional space. The collection frequency can be configured based on the scenario, but it is generally recommended to be no less than once every five seconds to ensure the continuity and real-time nature of the cluster status judgment.

[0023] Secondly, obtain temperature data from each vehicle's battery system. Battery temperature data can be read in real time by the temperature sensor in the vehicle's battery management system (BMS). This data includes cell temperature, battery pack case temperature, and thermal runaway warning status. This data is transmitted in degrees Celsius and uploaded in conjunction with the vehicle's unique identification and location data. Abnormal temperatures may indicate risks such as thermal runaway, partial short circuits, and cooling failures, and are key indicators in aggregated risk analysis.

[0024] State of Charge (SOC) data is also a key component of the data collection process. SOC reflects the ratio of the battery's current remaining charge to its full state, typically expressed on a scale of 0% to 100%. The collected SOC value is used to determine the vehicle's current charging behavior and battery load. Excessively high or low SOC levels can increase safety risks in large crowds, especially when continuous charging near full charge can easily lead to thermal anomalies.

[0025] Finally, the vehicle's operating status tag must be collected. This tag identifies the typical current state of the vehicle, including at least five types: "charging," "not charging," "standby," "offline," and "abnormal state." Abnormal states can be further categorized as thermal runaway alarms, BMS failures, communication interruptions, and more. This tag can be reported by the vehicle or intelligently determined by the platform based on temperature, SOC, and position changes. The operating status tag not only reflects the condition of the individual vehicle but also forms the basis for forming a weighted factor for clustered states.

[0026] The state acquisition module 101 encapsulates all of the above data in a unified structured format, including fields such as the vehicle's unique identifier, timestamp, geographic location, battery temperature, SOC value, and status tag, to form a standard data record. This data record is uploaded to the platform in real time via wireless communication (such as 4G, NB-IoT, WiFi, or LoRa) and stored in a dynamic state database or cache for subsequent simultaneous access and processing by the cluster identification module 102 and thermal risk prediction module 106.

[0027] To improve data collection efficiency and quality, the state collection module 101 can also be configured with data preprocessing functions, including location data denoising, temperature sampling averaging, state flag correction, and outlier removal. This module can also implement a time synchronization mechanism to ensure cross-device and cross-region data timing alignment, supporting global cluster evolution analysis within a sliding time window.

[0028] In summary, the status acquisition module 101 not only provides basic data support for the assessment of the aggregation status of new energy vehicles, but also provides executable input conditions for functional modules such as cluster identification, risk modeling and predictive response of the entire prevention and control system through real-time, high-dimensional and structured data output, ensuring that the system has an accurate, continuous and scalable data source.

[0029] The cluster identification module 102 is configured to identify clusters based on the operating status data using a density-based spatial clustering algorithm to obtain cluster characteristics of the clusters.

[0030] Cluster identification module 102 is used to identify groups of new energy vehicles clustered in specific areas and at specific times from the operating status data acquired by status acquisition module 101, and to extract the clustering characteristics of each group. Its core goal is to identify spatially dense and similarly operating status groups of vehicles, known as "clusters," from massive amounts of spatial location data, providing reliable input for subsequent risk modeling and response control.

[0031] In practical applications, the module first receives data such as the real-time location coordinates, battery temperature, state of charge (SOC), and operating status tags for each new energy vehicle. This data may be uploaded in real time by the vehicle's onboard system, charging station, ground positioning device, or communication module. After being standardized, it enters the cluster identification module. The module then analyzes the vehicles using a density-based spatial clustering algorithm. The basic idea of ​​this algorithm is that if a vehicle has a sufficient number of neighboring vehicles within a predetermined distance range, it is considered a "core point," from which a cluster can be formed.

[0032] For example, the system can set a fixed radius (such as 100 meters) and require at least five vehicles to be within it to be considered a cluster. This approach automatically adapts to the distribution of vehicles, eliminating the need to manually set the number of clusters. It can also effectively identify "isolated" vehicles—those that do not belong to any clusters—and label them as noise points.

[0033] After identifying multiple clusters, the cluster identification module extracts a series of key features from each cluster to describe its structure and state. First, the number of vehicles within the cluster is calculated to reflect the cluster's size. Then, based on the location of each vehicle, the physical area covered by the cluster is determined, for example, using a minimum enclosing rectangle to enclose all vehicles. Next, the number of vehicles is divided by this area to obtain the unit density of vehicles, which reflects the spatial congestion level of the cluster.

[0034] The cluster identification module also performs statistical processing on temperature data. For example, it calculates the average battery temperature of all vehicles within a cluster to determine the overall thermal load level in that area. The system also calculates the percentage of vehicles with temperatures exceeding a preset threshold. A higher percentage indicates a greater likelihood of thermal anomalies in that area. Similarly, the system calculates the percentage of vehicles with abnormal operating conditions, indicating possible risks within the cluster, such as operational failures or communication interruptions.

[0035] All of these features, including the number of vehicles, area occupied, unit density, average temperature, proportion of high-temperature vehicles, and proportion of vehicles in abnormal conditions, together constitute a complete description of the cluster. The cluster identification module outputs these feature data in a structured manner and transmits them to the subsequent regional risk modeling module 103.

[0036] To adapt to the dynamic urban traffic environment, cluster recognition modules typically run automatically at a fixed interval, refreshing the recognition results every 5 or 10 seconds, for example. The module can set a sliding time window to track the state of the same cluster at different points in time, identifying whether the cluster is expanding, contracting, or moving. Furthermore, the module can integrate post-processing mechanisms to merge or split clusters near their boundaries, preventing recognition fragmentation caused by positional errors and improving overall recognition stability and accuracy.

[0037] Through the above mechanism, the cluster identification module 102 can accurately and timely identify the aggregation behavior of new energy vehicles in high-density areas of the city, and output cluster data with physical characteristics and thermal state characteristics, providing clear and definite data support for the system's risk modeling and control strategy, ensuring that the system can realize intelligent aggregation situation identification capabilities for actual application scenarios.

[0038] Furthermore, the cluster identification module is specifically used to:

[0039] Construct a four-dimensional state space that combines the vehicle's geographic coordinates, battery temperature data, state of charge data, and vehicle operating status labels. This four-dimensional state space is normalized using a state similarity function guided by the state labels, providing a unified measurement basis for spatial distance calculations across different dimensional features.

[0040] In the four-dimensional state space, a dynamic density estimation mechanism is implemented. The dynamic density estimation mechanism dynamically adjusts the neighborhood search radius and minimum sample size of the clustering algorithm based on the proportion of high-temperature vehicles and density fluctuations of abnormal vehicles in the local space, making the clustering results more sensitive to potential thermal runaway areas;

[0041] After completing the preliminary clustering, a structural stability determination process is introduced for each cluster. Based on the fluctuation degree of the temperature gradient at the cluster boundary, the unstable boundary area is identified and boundary correction is performed to supplement the edge vehicles that are missed by the critical state during the clustering process.

[0042] The clustering characteristics of each cluster obtained are output, which include the number of vehicles, spatial coverage, vehicle density per unit area, average temperature value of the cluster, proportion of high-temperature vehicles, proportion of vehicles in abnormal conditions, and temperature fluctuation index at the cluster boundary, which are used to characterize the thermal risk sensitivity and evolution tendency of the cluster.

[0043] This embodiment first needs to make full use of the multi-source operating status data obtained by the status acquisition module. Specifically, the data collected by this module includes at least the geographic coordinates, battery temperature data, state of charge data, and operating status labels of each new energy vehicle. These data are both timely and differentiated, reflecting the distribution of vehicles in physical space, the thermal stability level of the electrochemical system, the remaining energy state, and the real-time operating status (for example: charging, not charging, abnormal, etc.). In order to achieve a more refined cluster identification mechanism that is more in line with the actual evolution of risks, the cluster identification module fuses these multi-dimensional data into a four-dimensional state space.

[0044] The establishment of this four-dimensional state space is not a simple splicing, but rather the introduction of a state similarity function guided by state labels, which unifies location, temperature, state of charge, and state labels into a measurement framework. This similarity function first normalizes the data of different dimensions. For example, the geographic coordinates are mapped to a unified unit according to the city scale, the temperature range is compressed to a standard interval of 0 to 1, the state of charge is converted into a relative charging rate or remaining capacity ratio, and the operating state labels are assigned weighting factors through numerical mapping, such as "abnormal state" is 1, "charging" is 0.75, and "idle state" is 0.5. In this way, each dimension of features can be made comparable and have a unified measurement basis in subsequent spatial distance or density estimation, thus avoiding the imbalance problem of one dimension of information dominating and other dimensions being ignored during clustering.

[0045] After constructing the four-dimensional state space, the cluster identification module will enter the core density estimation and clustering stage. Unlike the traditional DBSCAN or OPTICS algorithms that fix the neighborhood search radius and the minimum number of points, this implementation scheme adopts a dynamic density estimation mechanism to deal with the thermal risk sensitivity characteristics of new energy vehicle clusters. Specifically, in each candidate clustering area, the system first counts the proportion of high-temperature vehicles in the local space, as well as the density of vehicles with abnormal operating status labels. If the high temperature ratio in a certain area is significantly higher than that in the surrounding area, or there are multiple abnormal vehicles clustered, the system will automatically reduce the clustering neighborhood radius of the area and increase the clustering density threshold, making it easier to identify the core area of ​​risk clustering behavior. Conversely, for areas with stable conditions and slow temperature changes, the system can moderately relax the clustering conditions to improve recognition coverage and fault tolerance. This mechanism ensures that the system has higher recognition accuracy for thermal risk-sensitive areas, while avoiding excessive clustering of normal areas.

[0046] After completing the initial clustering, the system performs a structural stability assessment on each identified cluster to further improve recognition accuracy and cluster structure stability. This assessment focuses on cluster boundaries, specifically examining the temperature distribution gradient of vehicles at the edge of the cluster. If the temperature in the boundary area rises or falls sharply, or if the temperature fluctuation exceeds a set threshold, the system will determine that the boundary is structurally unstable. At this point, the module executes boundary correction logic to re-incorporate some critical vehicles at the boundary into the cluster, thereby compensating for edge omissions caused by cluster parameter settings and ensuring that the cluster fully reflects the true physical clustering situation.

[0047] Finally, cluster features will be extracted for each adjusted cluster. These features cover multiple dimensions, including: the number of vehicles in the cluster, which is used to reflect the scale of the cluster; the spatial range covered by the cluster, which can be calculated by the minimum circumscribed rectangle or convex hull method; the vehicle density per unit area, which reflects the degree of vehicle congestion; the average temperature value, which represents the overall thermal environment level; the proportion of high-temperature vehicles, which is used to indicate potential local thermal anomalies; the proportion of vehicles in abnormal states, which reflects the extensiveness of system anomalies; and the boundary temperature fluctuation index, which reflects the stability and evolution trend of the cluster structure. All these cluster features will be uniformly structured and output to provide complete, rich, and quantifiable input data support for the subsequent regional risk modeling module, so that the entire new energy vehicle cluster risk prevention and control system has the ability of dynamic and accurate identification, real-time thermal risk assessment, and intelligent intervention control. The following is the reference implementation code of the cluster identification module:

[0048] import numpy as np

[0049] import pandas as pd

[0050] from sklearn.cluster import DBSCAN

[0051] from scipy.spatial import ConvexHull, distance_matrix

[0052] # Sample data preparation (simulating vehicle status)

[0053] #Simulate vehicle data, including geographic coordinates, temperature, SOC, status labels (numeric mapping)

[0054] data = pd.DataFrame([

[0055] [30.1234, 120.5678, 45.0, 0.8, 'charging'],

[0056] [30.1235, 120.5679, 50.5, 0.9, 'abnormal'],

[0057] [30.1240, 120.5685, 42.0, 0.7, 'idle'],

[0058] [30.1300, 120.5700, 39.0, 0.6, 'idle'],

[0059] [30.1232, 120.5672, 49.0, 0.85, 'abnormal'],

[0060] [30.1236, 120.5676, 48.5, 0.8, 'charging'],

[0061] [30.1233, 120.5673, 46.0, 0.75, 'idle']

[0062] ], columns=['lat', 'lon','temp', 'soc', 'status'])

[0063] # Status label mapping

[0064] status_weight = {'abnormal': 1.0, 'charging': 0.75, 'idle': 0.5}

[0065] data['status_val'] =data['status'].map(status_weight)

[0066] # State similarity normalization function

[0067] def normalize(df, col, min_v=None, max_v=None):

[0068] if min_v is None:

[0069] min_v = df[col].min()

[0070] if max_v is None:

[0071] max_v = df[col].max()

[0072] return (df[col]- min_v) / (max_v - min_v + 1e-6)

[0073] data['lat_n'] =normalize(data, 'lat')

[0074] data['lon_n'] =normalize(data, 'lon')

[0075] data['temp_n'] =normalize(data, 'temp', 30, 60)

[0076] data['soc_n'] =normalize(data, 'soc', 0.0, 1.0)

[0077] data['status_n'] =data['status_val'] # Status has been mapped, use directly

[0078] # Construct four-dimensional state space: latitude and longitude + temperature + state

[0079] feature_vector = data[['lat_n', 'lon_n', 'temp_n', 'status_n']].values

[0080] # Dynamic density estimation mechanism (adjusting eps and min_samples)

[0081] high_temp_threshold = 47

[0082] abnormal_density = (data['status'] =='abnormal').sum() / len(data)

[0083] hot_density = (data['temp']>high_temp_threshold).sum() / len(data)

[0084] # Automatically adjust clustering parameters based on thermal risk sensitivity

[0085] if abnormal_density + hot_density>0.6:

[0086] eps = 0.02

[0087] min_samples = 2

[0088] else:

[0089] eps = 0.05

[0090] min_samples = 2

[0091] # Perform clustering (DBSCAN)

[0092] clustering = DBSCAN(eps=eps, min_samples=min_samples).fit(feature_vector)

[0093] data['cluster'] = clustering.labels_

[0094] # Clustering post-processing: structural stability analysis and boundary correction

[0095] clusters = []

[0096] for cluster_id in sorted(data['cluster'].unique()):

[0097] if cluster_id == -1:

[0098] continue # skip noise

[0099] cluster_data = data[data['cluster'] == cluster_id].copy()

[0100] coords = cluster_data[['lat', 'lon']].values

[0101] if len(coords)<3:

[0102] area = 0

[0103] boundary_std = 0

[0104] else:

[0105] hull = ConvexHull(coords)

[0106] area = hull.volume # used as two-dimensional area

[0107] boundary_idx = np.unique(hull.vertices)

[0108] boundary_temps = cluster_data.iloc[boundary_idx]['temp'].values

[0109] boundary_std = np.std(np.diff(sorted(boundary_temps))) # Temperature fluctuation gradient

[0110] # If the boundary temperature fluctuates too much, try to merge edge vehicles

[0111] if boundary_std>2.5:

[0112] # Calculate the Euclidean distance to the boundary points and try to find the missing vehicles

[0113] border_points = cluster_data.iloc[boundary_idx][['lat_n','lon_n']].values

[0114] other_points = data[data['cluster']== -1][['lat_n', 'lon_n']].values

[0115] dist = distance_matrix(other_points,border_points)

[0116] close_points_idx = np.where(np.min(dist,axis=1)<eps)[0]

[0117] new_indices = data[data['cluster']== -1].iloc[close_points_idx].index

[0118] data.loc[new_indices, 'cluster']= cluster_id

[0119] cluster_data = data[data['cluster']== cluster_id] # 更新

[0120] # Recalculate cluster features

[0121] updated_coords = cluster_data[['lat', 'lon']].values

[0122] if len(updated_coords)>= 3:

[0123] area = ConvexHull(updated_coords).volume

[0124] else:

[0125] area = 0.0001 # To prevent division by zero

[0126] vehicle_count = len(cluster_data)

[0127] density = vehicle_count / area

[0128] avg_temp = cluster_data['temp'].mean()

[0129] high_temp_ratio = (cluster_data['temp']>high_temp_threshold).sum() / vehicle_count

[0130] abnormal_ratio = (cluster_data['status']== 'abnormal').sum() / vehicle_count

[0131] cluster_summary = {

[0132] 'cluster_id': cluster_id,

[0133] 'vehicle_count': vehicle_count,

[0134] 'spatial_area': round(area, 5),

[0135] 'vehicle_density': round(density, 2),

[0136] 'avg_temp': round(avg_temp, 2),

[0137] 'high_temp_ratio': round(high_temp_ratio, 2),

[0138] 'abnormal_ratio': round(abnormal_ratio, 2),

[0139] 'boundary_temp_fluctuation': round(boundary_std, 2)

[0140] }

[0141] clusters.append(cluster_summary)

[0142] The regional risk modeling module 103 is used to calculate the risk score value of each cluster based on the cluster characteristics provided by the cluster identification module; the risk score value is obtained based on the vehicle density per unit area, the average temperature value of the cluster, the proportion of high-temperature vehicles, the proportion of vehicles in abnormal conditions and the charging load ratio of the target area.

[0143] The regional risk modeling module 103 performs a quantitative risk assessment on each cluster output by the cluster identification module 102. Its core function is to establish a computable risk scoring mechanism based on the statistical characteristics of the clusters, thus providing a basis for subsequent risk classification and response decisions. This module is based on a comprehensive modeling of multiple factors, including spatial distribution, vehicle status, and environmental loads, ensuring that the output score not only reflects the size and density of the current cluster but also accurately captures its potential thermal safety hazards.

[0144] In practice, the regional risk modeling module first receives basic statistical characteristics of clusters. These characteristics include the number of vehicles within each cluster, the spatial extent occupied, the average battery temperature, the proportion of vehicles with temperatures exceeding safety thresholds, the proportion of vehicles with abnormal operating conditions, and the current charging load ratio of the target area. Vehicle density per unit area is calculated as the ratio of the number of vehicles to the area occupied, reflecting the spatial compactness of the cluster. The average cluster temperature is a weighted average of the battery temperature data of all vehicles in the cluster, representing the overall thermal state of the cluster. The proportion of high-temperature vehicles refers to the proportion of vehicles in the cluster with temperatures exceeding a preset upper limit (e.g., 50°C), reflecting the potential risk of thermal runaway. The proportion of abnormal vehicles refers to the proportion of vehicles with abnormal status labels (e.g., communication failure, thermal warning, hardware failure, etc.). The charging load ratio of the target area is the ratio of the area's used charging power to its maximum loadable power, representing the energy load pressure in the environment.

[0145] The regional risk modeling module assigns a risk score value to each cluster based on these input factors. The scoring process can use a linear weighted model, in which each indicator has a corresponding importance weight. These weights can be obtained through training of historical accident samples, or they can be set empirically based on domain knowledge. Before the system is deployed, cluster samples can be collected based on multiple typical scenarios to record whether they have experienced thermal alarms or safety incidents. This is used as a sample label, and a set of stable and effective weight combinations is trained through methods such as least squares fitting or logistic regression. The calculation result of each score value corresponds to a numerical range, usually a real number, such as a value between 0 and 1 or 0 to 100. The higher the value, the higher the risk level.

[0146] To improve the interpretability and consistency of the scoring, the system normalizes all input data to ensure that the output score is not affected by absolute dimensions. For example, in vehicle density calculations, a normalization range can be determined based on the historical average and maximum values ​​for the current area. For temperature processing, a baseline temperature (e.g., 45°C) can be set as the critical point for adjusting segment weights. The resulting score reflects the overall risk level of each cluster at the current moment, providing direct input to the subsequent risk level determination module 104.

[0147] Furthermore, to adapt to diverse scenarios and dynamic environmental changes, the regional risk modeling module supports online updates of model parameters. For example, if the monitoring platform detects that certain high-scoring clusters have not triggered any abnormal events, or that low-scoring clusters have triggered thermal runaway alarms, the system can automatically adjust parameter weights to optimize the accuracy and sensitivity of the scoring mechanism.

[0148] The module's output is provided in a structured format, including at least the cluster number, score value, score timestamp, key parameter values ​​involved in the calculation, and risk explanations. These results are not only used by internal modules but can also be used for logging, trend analysis, or platform visualization, helping operations personnel intuitively understand the current regional thermal safety risk situation.

[0149] Through the above design and process, the regional risk modeling module 103 plays a key role in connecting the upper and lower levels of the system. It not only converts complex and multi-dimensional vehicle aggregation information into quantifiable risk signals, but also provides direct and clear input parameters for the scheduling and prediction modules, ensuring that the entire prevention and control system has good response efficiency and accuracy.

[0150] Furthermore, the regional risk modeling module is specifically used to:

[0151] The vehicle density per unit area and the average temperature value of each cluster in a continuous time period are differentially calculated to obtain the density change rate and temperature rise rate;

[0152] Perform sliding window statistics on the boundary temperature data of the cluster within a specified time period, extract the boundary temperature fluctuation amplitude, and use it to characterize the stability of the cluster structure;

[0153] According to the density change rate, temperature rise rate and boundary temperature fluctuation amplitude, the vehicle density per unit area, the average temperature value of the cluster, the proportion of high-temperature vehicles and the proportion of abnormal state vehicles are proportionally corrected;

[0154] The corrected vehicle density per unit area, average cluster temperature, proportion of high-temperature vehicles, proportion of vehicles in abnormal conditions, and charging load ratio of the target area are used together to calculate the risk score of each cluster.

[0155] In actual applications, it is often difficult to accurately capture the changing trend of risk situations by relying solely on the cluster characteristics at a certain moment. Therefore, this embodiment introduces the evolutionary characteristics of clusters to dynamically modify the scoring parameters to enhance the temporal responsiveness and structural stability identification capabilities of regional risk modeling, thereby achieving more accurate risk assessment.

[0156] First, during each round of risk scoring, the system extracts historical data showing the vehicle density per unit area and the average temperature of each cluster over a continuous time period. The vehicle density per unit area is calculated by dividing the number of vehicles in the cluster by the spatial area covered by the cluster, which can be calculated using the convex hull algorithm or the minimum enclosing rectangle method. The average temperature is the arithmetic mean of the battery temperatures of all vehicles within the cluster. The system then performs a differential calculation on these values ​​recorded at multiple consecutive time points—dividing the change in value between two consecutive time points by the time interval—to obtain the density change rate and temperature rise rate, respectively. These two parameters reflect the trend in aggregation strength and the rate of thermal accumulation evolution during the clustering process. For example, if a cluster's density per unit area increases rapidly and its temperature rise rate increases significantly over a short period of time, this indicates that the cluster may be rapidly evolving into a high-risk area.

[0157] Secondly, the system further extracts thermal stability indicators for the cluster boundary areas. To this end, sliding window statistical processing is performed on the vehicle temperature data at each cluster boundary. Boundary determination can be based on the distance of the vehicles in the cluster to the geometric center, selecting a certain proportion of samples close to the boundary. A fixed-length time window (such as 5 minutes or 10 minutes) is set. Within this window, the range, variance, or maximum gradient of the boundary vehicle temperature is calculated to quantify the temperature fluctuation amplitude. This fluctuation amplitude reflects whether there is significant thermal instability at the boundary. If the fluctuation is significant, there may be a risk that the cluster structure is about to break or expand, so the system should increase the level of attention to this cluster.

[0158] Combining the two sets of time evolution parameters mentioned above, the system uses them to make proportional corrections to the original scoring indicators. Specifically, the original scoring indicators include vehicle density per unit area, average temperature value of the cluster, proportion of high-temperature vehicles, and proportion of vehicles in abnormal conditions. If the density change rate of a cluster is high or the temperature rise rate is abnormally rapid, the scoring weights corresponding to its vehicle density per unit area and average temperature value can be appropriately increased, thereby enhancing the system's sensitivity to rapidly gathering hotspots. Similarly, if the boundary temperature fluctuation index exceeds the stability threshold, the weights of the proportion of high-temperature vehicles and the proportion of vehicles in abnormal conditions can be increased accordingly to reflect the potential structural risks of the area.

[0159] Finally, the system inputs the revised scoring index and the charging load ratio of the target area into the scoring calculation logic to comprehensively evaluate the risk level of each cluster. The scoring logic can be linearly summed by setting the weighted coefficients of each scoring factor, or a multi-level segmented judgment method can be used to ensure that the scoring results can simultaneously reflect the dynamic evolution characteristics of thermal, density and structural risks. The final output risk score value can not only accurately reflect the static risk status of the current cluster, but also has a certain foresight and trend judgment ability, providing a more reliable input data basis for the subsequent risk level determination module and response control scheduling module. Through the method described in this embodiment, high-sensitivity dynamic modeling and intelligent quantitative evaluation of new energy vehicle aggregation risks can be achieved, significantly improving the thermal safety management efficiency and precise intervention capabilities of new energy aggregation areas.

[0160] Here is an example to make it easier for those skilled in the art to understand. For example, in the actual operation of the new energy vehicle gathering prevention and control system, a city monitoring system identified a new energy vehicle cluster with typical gathering characteristics between 10 a.m. and 10:10 a.m. The real-time monitoring data of the cluster showed that at 10:00 a.m., there were 50 vehicles in the area, with a vehicle distribution area of ​​2,000 square meters and an average battery temperature of 45.0°C. By 10:05 a.m., the number of vehicles increased to 60, the distribution area shrank to 1,800 square meters, and the average temperature rose to 46.5°C. By 10:10 a.m., the number of vehicles further increased to 70, the coverage area shrank to 1,650 square meters, and the average battery temperature rose to 48.0°C.

[0161] Based on this time series data, the system first calculated the vehicle density per unit area for the cluster over three consecutive time slices, which were 0.025, 0.0333, and 0.0424 vehicles per square meter, respectively. The rate of change of density between two adjacent time slices was further calculated, yielding 0.00166 and 0.00182 vehicles per square meter per minute, respectively. Taking the average, the density change rate was approximately 0.00174 vehicles per square meter per minute. Simultaneously, the temperature change rate was also calculated over the same time period, with both changes occurring at 0.3°C per minute, indicating a trend of continued temperature increase within the cluster.

[0162] The system then focused on the cluster boundary area, analyzing temperature fluctuations in the outermost 10% of vehicles. Within the set sliding window, the temperature fluctuations of these vehicles were recorded to be 2.5°C, 3.0°C, 2.8°C, 3.1°C, 2.9°C, 3.2°C, and 3.0°C, respectively. The calculated average fluctuation of the boundary temperature was 2.93°C. Considering the system's set stability threshold of 2.5°C, this value clearly exceeds the stable range. The system therefore determines that the current cluster boundary is highly unstable, triggering a correction process for the scoring parameters.

[0163] Based on the cluster evolution characteristics, the original cluster characteristic values ​​were dynamically revised. The vehicle density per unit area increased by 10% due to rapid growth, from 0.0424 vehicles / square meter to 0.0466 vehicles / square meter. The average temperature increased by 5% due to a significant temperature rise trend, from 48.0°C to 50.4°C. The proportion of high-temperature vehicles, originally 30% (i.e., 21 out of 70 vehicles had battery temperatures exceeding 50°C), was revised up by 15% to 34.5% due to boundary fluctuations. The proportion of vehicles in abnormal conditions, originally 10%, was revised up by 15% to 11.5%. Furthermore, the system obtained a charging load ratio of 0.80 for the target area.

[0164] Finally, the system comprehensively evaluates the modified parameters and calculates a risk score using a pre-defined linear weighting method. The weights for each feature are: vehicle density per unit area: 0.2; average temperature: 0.25; proportion of high-temperature vehicles: 0.2; proportion of vehicles in abnormal conditions: 0.15; and charging load ratio: 0.2. The calculated results are as follows: vehicle density per unit area: 0.2 × 0.0466 = 0.00932; average temperature: 0.25 × 50.4 = 12.6; proportion of high-temperature vehicles: 0.2 × 0.345 = 0.069; proportion of vehicles in abnormal conditions: 0.15 × 0.115 = 0.01725; and charging load ratio: 0.2 × 0.8 = 0.16. Adding these values ​​together yields a final risk score of approximately 12.86 for this cluster.

[0165] This score is higher than the high-risk threshold set by the system (for example, 10.0), so the cluster will be automatically identified as a high-risk area by the system and trigger corresponding intervention measures, such as early warning notifications, enhanced thermal runaway monitoring, and charging power adjustment strategies.

[0166] Furthermore, the regional risk modeling module is further configured to:

[0167] Based on the spatial boundary morphology of each cluster, the position sequence of the boundary vehicles in a continuous time period is identified, and a boundary trajectory set is constructed, wherein the boundary trajectory set is used to preserve the geographical distribution path of the boundary vehicles over time;

[0168] Using the boundary trajectory set, the battery temperature of each boundary vehicle in the corresponding time slice is extracted to construct a boundary temperature sequence set with a time index. The boundary temperature sequence set is used to record the dynamic evolution process of the thermal state of the cluster boundary;

[0169] Based on the boundary temperature sequence set, the temperature variation amplitude, temperature rising rate and continuous temperature difference segment length in each window are extracted in a time sliding window manner to form a local temperature disturbance feature vector sequence;

[0170] The local temperature disturbance feature vector sequence is subjected to boundary directional aggregation, and the overall boundary temperature disturbance characteristic curve is constructed according to the arrangement order of each boundary segment in the spatial trajectory. The boundary temperature fluctuation amplitude is extracted based on its fluctuation amplitude as output to characterize the thermal stability risk of the cluster structure.

[0171] In the new energy vehicle aggregation prevention and control system, in order to more accurately assess the thermal stability risk of the cluster structure, the regional risk modeling module needs to further extract and analyze the dynamic evolution process of the thermal state of the cluster boundary.

[0172] First, to construct a representation of the thermal signature evolution of cluster boundaries, it is necessary to identify individual vehicles within each cluster's spatial boundary morphology within a specific time period. Boundary vehicle identification is not accomplished through static spatial distance judgment, but rather based on the topological structure of the cluster's outer contours after clustering. By calculating the outer contour envelope of the cluster and matching extreme points within the boundary neighborhood, vehicle trajectory points that meet the boundary determination criteria are selected as a set of candidate boundary points. Furthermore, temporal continuity must be considered. For each cluster, the boundary identification process is performed on multiple consecutive time slices to form a time-indexed set of boundary vehicle trajectory points. To preserve the evolution of the boundary over time, trajectories are further connected based on vehicle identification. The boundary positions of the same vehicle within different time slices are formed into a trajectory sequence, which is then constructed into a boundary trajectory set.

[0173] After obtaining the boundary trajectory set, it is necessary to extract the battery temperature data of these boundary vehicles in the corresponding time slices. Since the location of the boundary vehicles has the spatial outer edge attribute, their temperature changes are more sensitive to environmental changes and vehicle clustering effects. Therefore, the temperature data extracted from the boundary trajectory set can more effectively reflect the thermal state of the overall edge of the cluster. The specific operation is: according to the timestamp index order, for each time point in each boundary trajectory, the battery temperature value of the corresponding vehicle at that moment is obtained, and accumulated one by one to form a time-temperature mapping sequence with vehicle trajectory as the unit. In order to support the subsequent extraction of temperature disturbance features, these time-temperature sequences need to be uniformly stored in a boundary temperature sequence set with time index, so that the boundary thermal state of each cluster in each time slice has a clear structural expression and time relationship.

[0174] After the boundary temperature series are constructed, dynamic disturbance feature extraction is required. A sliding window approach is used to locally slice the temperature series along the time dimension. Within each window, three key disturbance feature indicators are sequentially extracted: the temperature variation amplitude, defined as the difference between the maximum and minimum temperatures within the window, characterizes the intensity of boundary temperature fluctuations within the window; the temperature rise rate, defined as the average increase in the rear-end value of the temperature series relative to the front-end value within the window, identifies trends in boundary heat accumulation; and the continuous temperature difference segment length, defined as the maximum time period within the window where the continuous temperature difference exceeds a preset fluctuation threshold, captures persistent warming. These three disturbance features collectively constitute a local temperature disturbance feature vector. The disturbance feature vectors of each sliding window form a sequence of local temperature disturbance feature vectors in chronological order, laying the foundation for the subsequent generation of the overall boundary disturbance pattern.

[0175] After extracting local disturbance features, the local disturbance feature vectors on all boundary trajectories need to be aggregated to construct an overall boundary temperature disturbance characteristic curve. This curve construction involves not only splicing the time series of feature vectors but also introducing a spatial directional sorting mechanism to ensure that the disturbance features of each boundary segment are combined according to the spatial arrangement order of the vehicles on the boundary. Spatial directional sorting is based on the geometric continuity of the boundary trajectories. By calculating the angle between the directional vectors of adjacent trajectory segments, the natural connection direction of the spatial path is determined, and an ordered disturbance sequence is constructed from the starting point to the end point of the boundary. The resulting boundary temperature disturbance characteristic curve is both directional and continuous, and can truly reflect the direction and trend of temperature fluctuations in space at the cluster boundary.

[0176] Finally, it is necessary to extract the key indicator representing the thermal stability risk of the cluster structure, namely the boundary temperature fluctuation amplitude, based on the constructed boundary temperature perturbation characteristic curve. This indicator is extracted by identifying the maximum temperature difference between the local peak and the trough in the complete boundary temperature perturbation characteristic curve, and combining the spatial span information to filter out abnormal isolated fluctuation points, extract representative continuous fluctuation amplitude segments, and then output the temperature difference value of this segment as the final boundary temperature fluctuation amplitude. This indicator can effectively identify thermal disturbances caused by cluster boundary instability or environmental coupling, and can serve as an important basis for the output of the regional risk modeling module to support the subsequent risk level determination, warning threshold calibration, and response scheduling strategy adjustment.

[0177] The risk level determination module 104 is configured to determine the risk level of the corresponding cluster according to the risk score value.

[0178] The risk level determination module 104 receives the risk score values ​​for each cluster output by the regional risk modeling module 103 and maps these scores into specific risk level labels based on a pre-defined multi-level risk classification standard. This provides clear and specific classification instructions for the subsequent response control scheduling module. The core function of this module is to convert continuous scoring results into executable classification signals and ensure that the boundaries between different levels are scientific, stable, and adaptable.

[0179] In practice, the risk level determination module first defines several risk level intervals, typically three or four levels, such as "low risk," "medium risk," and "high risk," or, in actual deployment scenarios, expanded to "normal," "warning," "high alert," and "emergency." Each level corresponds to a range of risk score values, with higher scores indicating a greater likelihood of thermal runaway or a safety incident. To establish these intervals, the module performs statistical modeling and threshold optimization based on extensive historical data. Methods such as ROC curve (Receiver Operating Characteristic) analysis and the maximum F1 value method are used to identify the optimal discriminant point, ensuring maximum classification accuracy when dividing scores between different levels.

[0180] During runtime, the risk level determination module compares the risk score of each cluster with the set threshold. If the score of a cluster is lower than the first threshold, it is classified as a low risk level; if the score is between the two middle thresholds, it is classified as a medium risk level; if the score exceeds the highest threshold, it is determined to be a high risk level. After each level determination is completed, a structured data record is generated, including the cluster number, the current score, the corresponding risk level, the determination timestamp, and the specific indicator item that triggered the level change. This record will be synchronously transmitted to the response control scheduling module to trigger the decision-making and execution of the flow control, guidance, or linkage strategy.

[0181] To account for the dynamic nature of the actual operating environment, the risk level determination module also supports an adaptive threshold adjustment mechanism. For example, if the overall system score is generally high during a specific time period or in a specific area, but no actual incidents have occurred, the high-risk threshold can be automatically raised to avoid misjudgments. Conversely, in areas where incidents have occurred or if a deterioration in the external thermal environment is detected, the threshold can be temporarily lowered to increase sensitivity. Furthermore, the module can also access the output of the thermal risk prediction module 106 to pre-adjust the current risk level determination boundary, ensuring that risk level determinations are more accurately aligned with future trends.

[0182] To ensure the consistency and stability of system responses, the risk level determination module uses a sliding time window mechanism to update the level, avoiding frequent triggering of level changes due to short-term fluctuations in the score value. For example, if the score value of a cluster exceeds the high-risk threshold in two consecutive cycles, the system will determine it as a persistent high-risk cluster and trigger a high-priority response. At the same time, to prevent delayed identification, the module also sets a fast response channel. Once the score value instantly exceeds the limit threshold, it will be immediately marked as an emergency level and notify the platform to implement emergency strategies such as power outages and dispersal.

[0183] To sum up, the risk level determination module 104 not only serves as a bridge from continuous risk score values ​​to graded response control signals, but also ensures the accuracy, stability and responsiveness of risk assessment results through a multi-level threshold system, adaptive optimization mechanism and time consistency constraints, thereby providing reliable protection for risk management in new energy vehicle aggregation scenarios.

[0184] Furthermore, the risk level determination module is specifically used to:

[0185] Obtaining risk score values ​​output by the regional risk modeling module for the same cluster in multiple consecutive time slices, and constructing a score time series in chronological order;

[0186] Based on the score time series, the incremental value of the risk score between each time slice and its average change rate are calculated to characterize the short-term score growth trend of the cluster;

[0187] Based on the short-term score growth trend, combined with the current score and static cluster characteristics including vehicle density per unit area and cluster average temperature, the risk score change range within a preset time window is predicted, and the future score prediction value range is output;

[0188] Perform interval overlap analysis on the future score prediction value interval and the static risk level threshold to identify whether there is an early warning signal of a trend crossing the high risk threshold;

[0189] When the early warning signal is established, the current score value and the upper limit of the future score prediction value range are combined to construct a trend reinforcement factor and correct the current score value before outputting the final score result;

[0190] The final scoring result is compared with the risk level determination rule to obtain the current risk level corresponding to the cluster.

[0191] In this embodiment, the risk level determination module undertakes the key task of converting the risk score value into a specific and identifiable risk level, and by introducing a time series trend analysis mechanism, the risk determination has dynamic evolution characteristics. The module first needs to obtain the risk score value output by the regional risk modeling module for the same cluster in multiple consecutive time slices. Generally speaking, the score value is calculated based on a unified cluster identification and is updated at fixed time intervals. Therefore, by collecting the score results of the most recent several time slices, a complete score time series can be constructed. The series is arranged in chronological order and reflects the risk evolution trajectory of the cluster in the recent period.

[0192] The system then performs a short-term growth trend analysis on this score time series. Specifically, it calculates the difference between scores in adjacent time slices to obtain the incremental risk score value at each moment. Based on this, it further calculates the average of all incremental scores to obtain the average risk score growth rate for the current cluster over the most recent period. This growth rate effectively quantifies the rate of change in risk scores, thereby characterizing the upward risk trend of the cluster and providing support for subsequent forecasts.

[0193] Next, the system constructs an input feature vector for future score predictions by combining the score value of the current time slice with several static characteristic indicators of the cluster, such as vehicle density per unit area and average cluster temperature. Based on a collaborative analysis of score growth trends and static characteristics, the system uses a regression algorithm or interval growth model to predict the range of possible risk score fluctuations within a preset time window and outputs this as a range of future score predictions. This range can be expressed as upper and lower limits to reflect the fluctuation boundaries of the score within that time window.

[0194] To determine whether future score evolution will trigger a change in risk level, the system incorporates an interval overlap analysis mechanism. Here, static risk level thresholds are preset as fixed segmented intervals, with low, medium, and high risk corresponding to different score ranges. When the predicted future score interval partially or completely overlaps with the high-risk threshold interval, the system identifies this as an early warning signal of a trend toward crossing the high-risk threshold. This signal is used to proactively identify potential sudden increases in risk within a cluster.

[0195] If the early warning signal is confirmed, the system will further construct a trend reinforcement factor using the current score and the upper limit of the future score prediction range. This factor, through a certain functional relationship, reflects the potential transition trend of the risk score and corrects the current score. This correction method can use linear amplification, exponential weighting, or threshold-driven methods to adjust the current score to a final score that better reflects the trend warning effect, ensuring the system's ability to respond promptly when it perceives evolving trends.

[0196] Finally, the system inputs the final scoring results into the risk level determination rules for comparison. These rules typically employ a multi-segment scoring system, determining the risk level of the cluster by the range within which the score falls. For example, if the score falls within a certain high-risk threshold, the cluster is classified as high risk. This final determination constitutes the output of the risk level determination module and provides direct guidance for downstream risk response controls.

[0197] Through this approach, the risk level assessment module not only makes static assessments based on the current score, but also integrates score trends and future predictions, creating an evolving and dynamic risk assessment mechanism. This makes the system proactive and highly sensitive to the risk of heat buildup from new energy vehicles. This mechanism effectively enhances the system's ability to identify and respond to sudden risks, breaking through the traditional approach of making assessments based solely on a single moment's score.

[0198] In a certain city center, the new energy vehicle monitoring system identified a high-density cluster, designated C1, through its cluster identification module. This cluster, identified at 3:00 PM on June 1, 2025, consisted of 47 vehicles. Preliminary characteristics included a vehicle density of 16.5 vehicles per thousand square meters, an average cluster temperature of 50.2°C, a 19.1% proportion of vehicles in high-temperature conditions, a 12.8% proportion of vehicles in abnormal conditions, and a charging load ratio of 0.72. The regional risk modeling module output risk scores for C1 for six consecutive 5-minute time slices between 3:00 PM and 3:25 PM: 0.42, 0.46, 0.51, 0.57, 0.64, and 0.72, respectively.

[0199] The risk level determination module first constructs a score time series S = [0.42, 0.46, 0.51, 0.57, 0.64, 0.72] and arranges it in chronological order. The system then performs a differential analysis on adjacent score values ​​to obtain a score increment series ΔS = [0.04, 0.05, 0.06, 0.07, 0.08]. The average of this series is calculated, yielding an average risk score growth rate of 0.06. This value serves as a quantitative indicator reflecting the growth trend of the current cluster risk score over a short period of time.

[0200] Next, the system obtains static cluster features at the current time (3:25 PM), including parameters such as a vehicle density per unit area of ​​16.5, an average cluster temperature of 50.2°C, a proportion of high-temperature vehicles of 19.1%, and a proportion of vehicles in abnormal conditions of 12.8%. Using a pre-defined score growth trend prediction model (which can be a linear or nonlinear regression function trained based on historical regression samples), the score growth rate of 0.06 is combined with these static features to predict the possible risk score range within a future time window (set to the next 15 minutes, from 3:30 PM to 3:45 PM). For example, according to the model output, the predicted score value for C1 within the future window is in the range [0.74, 0.86].

[0201] After obtaining the score prediction interval, the system compares it with the statically set risk level threshold. The high-risk threshold is set at 0.80. The system performs an interval overlap analysis between the prediction interval [0.74, 0.86] and the threshold interval [0.80, 1.00], finding a significant overlap between the two. The upper limit of the prediction exceeds the high-risk threshold, and the system records a warning signal indicating that the cluster is trending across the risk level.

[0202] Once the early warning signal is confirmed, the system constructs a trend enhancement factor. The weight parameter for the trend enhancement factor is set to α = 0.75. The current score, Scurrent = 0.72, and the upper limit of the forecast interval, Spred = 0.86, are weighted to calculate the revised score, Sfinal = α × Scurrent + (1 – α) × Spred = 0.75 × 0.72 + 0.25 × 0.86 = 0.755. This value serves as the final score, integrating the current state and future forecasts.

[0203] Finally, the system compares Sfinal = 0.755 with the risk level rules. The risk level rules are as follows: a score less than 0.6 is low risk, 0.6 to 0.8 is medium risk, and 0.8 and above is high risk. Because 0.755 falls at the upper limit of the medium-risk range and is predicted to evolve toward high risk, the system ultimately determines C1 as a critical medium-high risk state and pre-classifies it as high risk through a trend reinforcement mechanism. This level is then output to the intervention decision module to trigger regional charging restrictions and scheduling optimization.

[0204] Furthermore, the risk level determination module is further configured to:

[0205] The future score prediction value interval is decomposed into three indicators: the lower limit value, the upper limit value and the interval span. Based on the short-term score growth trend of the score time series, the distance between the interval span and the current score value is calculated to represent the fluctuation range of the possible future score growth;

[0206] According to the fluctuation range, all level cutoff values ​​greater than the current score value in the static risk level threshold are retrieved, a set of cutoff values ​​with overlapping values ​​is screened out, and the score difference interval corresponding to the set of cutoff values ​​is output;

[0207] Compare the score difference interval with the fluctuation range, calculate the coverage ratio of each cutoff value within the fluctuation range, and use the ratio as a trend crossing strength indicator to determine whether the future score interval has a tendency to migrate to a high-risk level;

[0208] When the trend crossing strength indicator reaches a preset judgment threshold, an early warning signal is generated.

[0209] In the new energy vehicle gathering prevention and control system, the risk level determination module analyzes the future score prediction value range and performs a series of processing steps to determine whether to generate an early warning signal. First, after obtaining the future score prediction value range, the module does not directly use the interval information as a whole to make a judgment. Instead, it breaks down the interval into three indicators: a lower limit value, an upper limit value, and an interval span. The lower limit value and the upper limit value represent the possible minimum and maximum values ​​of the future estimated score value, respectively, while the interval span is the difference between the upper and lower limits. This disassembly process ensures that subsequent processing can identify risk trends based on a more detailed structure.

[0210] Next, the module obtains the short-term score growth trend for the score time series corresponding to the score interval. This trend is calculated by calculating the change in score values ​​between adjacent time slices, extracting information about the magnitude and speed of score growth over a continuous period of time. Based on this short-term score growth trend, the module further calculates the numerical distance between the interval span and the current score value to form a fluctuation range metric. This numerical distance is used to quantify the extent to which the predicted score value interval may shift upward under the current state, reflecting the potential magnitude of score growth.

[0211] After the aforementioned fluctuation range calculation is complete, the module enters the matching process for the static risk level threshold. The system presets multiple static risk level thresholds as the boundaries for risk level division. The module iterates over all risk level cutoffs above the current score value, compares each cutoff value, and determines whether the cutoff value overlaps with the future score prediction range. The judgment is based on whether the upper and lower limits of the future score prediction range overlap the cutoff value. If so, it indicates that the risk level threshold may be crossed by the score value in the future.

[0212] For all overlapping cutoff values, the module proceeds to compare the score difference interval with the fluctuation range. Specifically, the score difference interval refers to the difference between the cutoff value and the current score value, and the fluctuation range is the difference between the upper limit of the score interval and the current score value. The module calculates the ratio of the score difference interval to the fluctuation range to determine the proportion of the cutoff value covered by the fluctuation range. This ratio is the trend crossing strength indicator. The calculation of this ratio must not only ensure the consistency of the upper and lower limits, but also unify the data source and time synchronization of the score values ​​to avoid inconsistencies caused by differences in scoring time.

[0213] After calculating the trend crossing strength index for all thresholds, the module compares each threshold's result with the system's preset threshold to determine whether the triggering condition for an early warning is met. If the trend crossing strength index for any threshold exceeds the threshold, it indicates that the score is trending toward crossing that threshold, and the module generates an early warning signal. This signal is then used as the status indicator for the current cluster and is subsequently used by the scheduling and guidance modules.

[0214] The response control scheduling module 105 is used to execute flow control and path guidance measures according to the risk level and provide response results to the external platform.

[0215] The response control scheduling module 105 is responsible for implementing intervention control strategies based on the cluster risk level output by the risk level determination module 104. This strategy mitigates or avoids safety risks such as thermal runaway, fire spread, and regional load overload caused by the concentration of new energy vehicles. This module not only includes the strategy formulation logic but also facilitates linkage and information exchange with terminal execution devices and external city management platforms. It is a key component for achieving closed-loop control and practical deployment of the system.

[0216] When the risk level determination module outputs that a certain cluster is at a medium or high risk level, the response control scheduling module first starts the current limiting control mechanism. Specifically, the system calculates the acceptable maximum vehicle load threshold for the area based on the spatial boundaries of the area where the cluster is located and the current number of vehicles. If the actual number of vehicles exceeds the threshold, the module will issue an access restriction instruction to the charging platform or parking management system, prohibiting new vehicles from entering the specified range of the area. This current limiting instruction can be directly applied to the charging pile, gate system or parking lot management system through an API interface or communication protocol, giving priority to limiting the access of vehicles with high temperature, high SOC or long-term residence. At the same time, the current limiting strategy has the ability to dynamically adjust. The system will periodically update the access upper limit value according to changes in the cluster status to avoid resource waste caused by long-term rigid thresholds.

[0217] If a high-risk level is determined, the response control and scheduling module will also simultaneously activate the route guidance strategy. The core purpose of this strategy is to promptly divert vehicles already within or approaching the risk area to a relatively safe area. The system automatically generates the optimal route based on factors such as the road structure, number of available parking spaces, regional temperature distribution, and traffic density within the current area. Guidance information is transmitted to the vehicle owner or vehicle control system via multiple terminals, including the on-board T-Box module, mobile application (APP), navigation terminal, and charging station display screen, prompting them to immediately leave the current area and instructing them to proceed to a recommended alternative area for parking or charging. For vehicles not connected to the network, the system can also communicate with on-site guidance screens or the broadcast system to provide regional notifications.

[0218] The response control and dispatch module not only includes current limiting and guidance functions, but also supports multi-level linkage with other subsystems. Upon a high-level risk assessment, the module can proactively send a linkage signal to the fire protection system, pre-activating ventilation or sprinkler systems to reduce local heat accumulation. It can also control the lighting system to activate a high-brightness flashing mode to alert personnel. It can also call upon the regional monitoring system's cameras to perform AI video analysis on the core area of ​​the cluster, identifying visual risk signals such as smoke, abnormal behavior, or hardware damage, and providing feedback to the monitoring backend.

[0219] Furthermore, the response control and dispatch module transmits execution results back to the system platform in real time and simultaneously uploads them to external interfaces such as the city management platform, energy dispatch platform, or emergency command center. Uploaded content includes the effectiveness of flow control, the implementation of vehicle diversion routes, the response status of linked equipment, and cluster trends, providing regulatory authorities with a complete response log and a basis for subsequent disposition. All execution actions and status changes are recorded in the system as an event chain and can be exported as operation logs, charts, or alarm reports, meeting the traceability and visualization requirements of city-level risk management.

[0220] To ensure high reliability, the responsive control scheduling module supports multi-threaded scheduling and priority queue management during deployment. This automatically schedules resources and allocates channels in the event of multiple clusters or sudden emergency gatherings, prioritizing control requests in the most critical areas based on risk level. Furthermore, to avoid false triggering and resource waste, the module incorporates a delayed confirmation mechanism and a minimum execution interval, ensuring that control actions are sufficiently responsive after actual demand is triggered without being executed excessively frequently.

[0221] Through the above mechanism, the response control scheduling module 105 realizes a closed-loop logic from risk identification to on-site control. It can not only respond in real time based on the current risk level, but also cooperate with the prediction module to deploy early warning measures in advance, so that the entire system has strong dynamic intervention and linkage control capabilities, thereby effectively ensuring the operational safety and urban management efficiency in scenarios with dense new energy vehicles.

[0222] The thermal risk prediction module 106 is used to construct a time series feature data set based on a sliding window according to the operating status data, and use a pre-trained long short-term memory neural network model to predict the potential risk score trend of each target area within a preset time range in the future; the predicted potential risk score trend is output to the risk level determination module and the response control scheduling module to execute the current limiting or guidance strategy in advance.

[0223] The thermal risk prediction module 106 is designed to further predict future risk trends in areas where new energy vehicles are concentrated, building on the system's current state monitoring and risk level assessment. This provides a scientific basis for the response control and scheduling module to proactively deploy flow control or guidance strategies. Based on time series modeling methods and combined with artificial intelligence algorithms, particularly the long-short-term memory (LSTM) neural network architecture, this module effectively captures the dynamic characteristics of regional aggregation over time and is suitable for addressing the nonlinearity, hysteresis, and short-term mutations that exist in the aggregation of new energy vehicles.

[0224] In the specific implementation process, the thermal risk prediction module first continuously receives the operating status data from the status acquisition module 101, and performs sliding window arrangement with time as the main axis. The so-called sliding window means that the system selects a time series of fixed length in each time step as the input sample for the current prediction. For example, it sets a step length of every 5 minutes, takes the data of the past 30 minutes to form a window, and advances the window by 5 minutes each time to form a continuous time series data set. The feature items contained in each window include but are not limited to the total number of vehicles in the target area, the vehicle density per unit area, the average battery temperature, the proportion of high-temperature vehicles, the proportion of vehicles in abnormal conditions, the charging load ratio of the current area, the temperature rise rate, etc. These feature items are combined into input vectors in a structured manner, arranged in chronological order to form training samples, and input into the trained prediction model.

[0225] This module uses LSTM as its primary prediction structure. LSTM is a type of recurrent neural network that effectively retains long-term dependency information in time series data, avoiding the vanishing or exploding gradient problems that can occur with traditional neural networks when processing sequences. The internal structure of LSTM includes an input gate, a forget gate, and an output gate. These gating mechanisms dynamically select which historical information to retain and which to discard, thereby maintaining the model's memory capacity while increasing its sensitivity to new state changes. The system establishes a corresponding LSTM model instance for each target area and completes offline training based on the historical operating data of the deployment area. During the training phase, the system uses the actual aggregation state and subsequent risk score values ​​for a known time period as training labels, optimizes the model parameters through backpropagation, and ultimately forms model weights that can autonomously predict future risk trends.

[0226] After deployment, the thermal risk prediction module can call the model in real time during system operation. In each prediction cycle, the module extracts time series features from the current sliding window, inputs them into the LSTM model, and outputs the prediction results for one or more future time steps. The prediction results are usually presented as continuous risk score values, indicating the risk score trend of the target area in the next 5 minutes, 10 minutes, or 15 minutes. The prediction results can not only be used for trend display, but can also be linked with the risk level determination module to compare the predicted score value with the determination threshold. If the future trend shows that a high risk is approaching, the system can mark the cluster as "predicted high risk" in advance, thereby triggering the response control scheduling module to intervene in advance and implement measures such as flow control, path guidance, or warning issuance.

[0227] To enhance the stability and reliability of predictions, the heat risk prediction module also incorporates weighted corrections based on the measured scores at the current time point. This uses a combination of the current and predicted scores as the final reference, while also setting confidence intervals to determine whether the predictions significantly deviate from the current situation. As the model continues to run, the system also periodically updates the model weights through online learning to adapt them to seasonal changes, regional characteristics, and user behavior patterns.

[0228] Ultimately, the thermal risk prediction module outputs structured data for use by other modules in the system and is simultaneously recorded on the risk management platform for trend analysis, historical review, and strategy evaluation. The prediction results are not only visualized as time series curves but can also be overlaid on regional maps to form heat maps, highlighting areas that may pose risks in the future and guiding managers and the scheduling platform to take proactive intervention measures.

[0229] Through this combination of time series modeling and LSTM intelligent prediction, the thermal risk prediction module 106 gives the system the ability to actively warn, so that the identification of new energy vehicle aggregation risks is no longer limited to the current situation level, but can achieve dynamic prediction extending into the future, providing key technical support for building a more efficient, smarter and safer urban new energy operating environment.

[0230] In order to facilitate those skilled in the art to better understand and implement the LSTM model used in the thermal risk prediction module, a specific example is provided below.

[0231] For example, the LSTM model consists of three main parts: the input layer, the LSTM processing layer, and the output layer.

[0232] The input layer receives multidimensional feature data collected continuously within a time window. The data for each time step includes seven features: total number of vehicles, vehicle density per unit area, average cluster temperature, proportion of high-temperature vehicles, proportion of vehicles in abnormal conditions, regional charging load ratio, and average temperature rise rate. Assuming the time window length is set to six time steps, corresponding to the past 30 minutes (5 minutes per step), the input layer receives a two-dimensional matrix of shape 6×7, where 6 represents the number of time steps and 7 represents the feature dimension of each step.

[0233] The output of the input layer is directly passed to the LSTM processing layer, which consists of a standard single-layer LSTM structure. The number of units can be set based on computing power, with 32 hidden units recommended. The LSTM processing layer progressively reads each time step of the input sequence and updates its internal state vector and output state at each moment. The final output of this processing layer is a one-dimensional vector of length 32, representing the representative state representation extracted by the model after comprehensively considering the characteristics of the historical time series.

[0234] The output layer receives the final output vector from the LSTM processing layer and maps it into a scalar through a fully connected (dense) neural network layer, representing the predicted risk score for one time step in the future (e.g., 5 minutes). During deployment, this score is fed into the risk level determination module and compared with a preset threshold to help determine whether to implement proactive control measures.

[0235] This model can be implemented using mainstream deep learning frameworks (such as TensorFlow or PyTorch). Model parameters are trained using historically annotated datasets, using the mean squared error (MSE) as the loss function. The Adam algorithm is recommended as the optimizer. During the training phase, a labeled sample sequence (the input is time series features, and the label is the actual future rating) is required to optimize the model weights in a supervised manner. During the inference phase, the model runs every prediction period (e.g., every 5 minutes), continuously outputting the rating prediction results for the next time step.

[0236] This structure has good real-time performance and deployment efficiency while ensuring prediction accuracy, and is suitable for dynamic thermal risk prediction of urban-level new energy vehicle aggregation trends.

[0237] Furthermore, the thermal risk prediction module is specifically used to:

[0238] Based on the operating status data, a multidimensional time series feature dataset is constructed, which includes the vehicle's geographic location, battery temperature, state of charge, and vehicle operating status labels, and the feature dataset is segmented in chronological order using a sliding window method;

[0239] A long short-term memory neural network model with a dynamic attention mechanism is used to encode the time series features within each sliding window to obtain a hidden state vector sequence, which is used to capture the temporal patterns and spatial evolution trends of vehicle aggregation behavior.

[0240] Generate a potential risk score trend for each target area within a preset future time range based on the spatial distribution of the hidden state vector sequence and its relationship with the cluster boundary structure, and output the potential risk score trend in the form of a structured tensor;

[0241] The output in the form of structured tensors is spatially mapped and converted into a regional-level heat map image format according to the geographic grid encoding rules of the target area. The image is then projected onto the target area map to form a predictive high-risk cluster area layer.

[0242] The risk score trend of each grid in the high-risk cluster area layer is output to the risk level determination module and the response control scheduling module to complete the scheduling planning of flow control, vehicle guidance and load distribution strategies in the area in advance.

[0243] In the new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling described in the present invention, the thermal risk prediction module is the core component of the prediction and early warning functions of the entire system. Its function is to analyze historical and current new energy vehicle operating status data to construct a prediction model with time dependence and spatial evolution laws, so as to judge the thermal risk evolution trend in the future time period in advance, and visualize the prediction results onto the target area map to assist in the pre-identification and intervention planning of high-risk gathering areas.

[0244] The thermal risk prediction module first uses operating status data as its foundation, including vehicle geographic location, battery temperature, state of charge, and vehicle operating status labels. This data is indexed by timestamps and constitutes a continuous raw time series dataset. To effectively analyze this time series data, the module adopts a sliding window processing strategy. Specifically, a fixed-length time window is selected and the entire raw time series dataset is sequentially traversed according to a set sliding step size. The data segments within each window constitute an independent subsequence, which contains the four-dimensional features of each vehicle at multiple time points: geographic coordinates, battery temperature, state of charge, and operating status label. Because the dimensions and ranges of different features vary significantly, after completing the sliding window segmentation, the system first normalizes all features to a unified scale, facilitating input encoding for the subsequent neural network model.

[0245] After feature normalization, the time series data within each sliding window is fed as input into a long-short-term memory neural network model regulated by a dynamic attention mechanism. This model, built on the basic structure of a long-short-term memory (LSTM) network, consists of multiple stacked layers of LSTM units. Each layer receives the state at the previous moment as input and outputs the hidden state vector at the current moment. A dynamic attention mechanism is introduced to enhance the model's responsiveness to critical moments and vehicle states. Within each time window, the attention mechanism dynamically evaluates the importance of different time slices and feature dimensions and weights the hidden states. This weighting strengthens the representation of areas with severe high-temperature fluctuations, vehicles with abnormal charge states, and nodes with abnormal operating states in the hidden state vector, ensuring that the model focuses on signal information that plays a key role in the evolution of thermal risks.

[0246] After processing each window, the model outputs the corresponding latent state vector sequence, which carries the evolutionary trajectory of the vehicle state within the entire time window. To transform the temporal encoding results into a spatially interpretable thermal risk trend output, the system further analyzes the clustered distribution characteristics of the latent state vectors in the spatial dimension. Combined with the cluster boundary structure information provided by the cluster identification module, it determines whether the current latent state vectors form a high-risk trend cluster. If the latent states are densely concentrated in a certain spatial area and contain a large number of vehicles in high-temperature or high-risk conditions within the corresponding time period, the area will be marked as a potential risk hotspot.

[0247] After obtaining the hidden state sequences of all sliding windows, the system statistically aggregates all time windows for each target region, calculates the potential risk score trend for that region within a preset future time period, and organizes this data into a structured tensor. This structured tensor is constructed along both spatial and temporal dimensions. The spatial dimension is determined by the target region's grid encoding rules, with each grid corresponding to a fixed geographic location. The temporal dimension is arranged sequentially according to the sliding window's time segmentation, ensuring that the tensor value for each grid reflects the predicted risk score for that geographic unit within different time slices. This structure not only preserves spatial variability but also presents temporal evolution trends, facilitating use in subsequent modules.

[0248] After the potential risk score trend in the form of the above-mentioned structured tensor is output, the system performs spatial mapping processing. Based on the pre-constructed target area map and geographic grid division scheme, each tensor unit is mapped to a grid block in the actual map. The system maintains the unique association between the tensor value and the geographic grid in each mapping process, that is, each data point in the tensor can only correspond to a clear geographic location unit on the map. After the mapping is completed, the system generates a regional-level heat map image format, where each pixel block represents a spatial unit and uses a color gradient to visually represent the future risk score change trend of the unit. The heat map can be updated in real time and dynamically superimposed on the map visualization interface to achieve dynamic display of the predictive high-risk cluster area layer.

[0249] After generating the high-risk cluster area layer, the system further processes the risk score trends corresponding to each grid in the layer and transmits them as prediction results to the risk level determination module and the response control scheduling module. The risk level determination module uses this score trend result to determine whether a grid's risk score is expected to increase significantly and exceed the high-risk threshold in the future time period. If so, the corresponding cluster is upgraded to a higher risk level in advance, triggering the early warning response mechanism. The response control scheduling module uses this score trend prediction result, combined with the actual vehicle location and current load status, to pre-determine flow control strategies, route guidance strategies, and load distribution plans, achieving proactive intervention and dynamic optimization control of regional thermal risks.

[0250] Through the above approach, the thermal risk prediction module of the present invention not only constructs a complete time series feature processing and deep neural prediction path, but also realizes a highly integrated process from prediction to map layer display through the enhanced representation of spatial structure, ensuring the feasibility, timeliness, and foresight of the system in engineering deployment. At the same time, the dynamic attention mechanism and structured spatial tensor output introduced in the module make the prediction process highly sensitive and interpretable, significantly superior to traditional static regression models and solutions based on single-point scoring prediction, and improving the recognition accuracy and response efficiency of the evolution process of the thermal risk of new energy vehicles.

[0251] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling, characterized by: include: A status acquisition module is used to collect the operating status data of new energy vehicles in the target area, wherein the operating status data includes the vehicle's geographic coordinates, battery temperature data, state of charge data, and vehicle operating status tags; A cluster identification module is used to identify clusters based on the operating status data using a density-based spatial clustering algorithm to obtain cluster characteristics of the clusters; A regional risk modeling module, configured to calculate a risk score value for each cluster based on the cluster characteristics provided by the cluster identification module; The risk score is obtained based on the vehicle density per unit area, the average temperature of the cluster, the proportion of high-temperature vehicles, the proportion of vehicles in abnormal conditions, and the charging load ratio of the target area; A risk level determination module, configured to determine the risk level of the corresponding cluster according to the risk score value; A response control scheduling module is used to execute flow control and path guidance measures according to the risk level and provide response results to the external platform; A thermal risk prediction module is configured to construct a sliding window-based time series feature dataset based on the operating status data, and use a pre-trained long-short-term memory neural network model to predict the potential risk score trend of each target area within a preset future time range; output the predicted potential risk score trend to the risk level determination module and the response control scheduling module to implement current limiting or guidance strategies in advance; The cluster identification module is specifically used for: Construct a four-dimensional state space that combines the vehicle's geographic coordinates, battery temperature data, state of charge data, and vehicle operating status labels. This four-dimensional state space is normalized using a state similarity function guided by the state labels, providing a unified measurement basis for spatial distance calculations across different dimensional features. In the four-dimensional state space, a dynamic density estimation mechanism is implemented. The dynamic density estimation mechanism dynamically adjusts the neighborhood search radius and minimum sample size of the clustering algorithm based on the proportion of high-temperature vehicles and density fluctuations of abnormal vehicles in the local space, making the clustering results more sensitive to potential thermal runaway areas; After completing the preliminary clustering, a structural stability determination process is introduced for each cluster. Based on the fluctuation degree of the temperature gradient at the cluster boundary, the unstable boundary area is identified and boundary correction is performed to supplement the edge vehicles that are missed by the critical state during the clustering process. Output the cluster characteristics of each cluster obtained. The cluster characteristics include the number of vehicles, spatial coverage, vehicle density per unit area, average cluster temperature, proportion of high-temperature vehicles, proportion of abnormal vehicles, and cluster boundary temperature fluctuation index, which are used to characterize the thermal risk sensitivity and evolution tendency of the cluster. The proportion of high-temperature vehicles refers to the proportion of vehicles with temperatures exceeding 50°C in the cluster.

2. The new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling according to claim 1 is characterized in that: The regional risk modeling module is specifically used to: The vehicle density per unit area and the average temperature value of each cluster in a continuous time period are differentially calculated to obtain the density change rate and temperature rise rate; Perform sliding window statistics on the boundary temperature data of the cluster within a specified time period, extract the boundary temperature fluctuation amplitude, and use it to characterize the stability of the cluster structure; According to the density change rate, temperature rise rate and boundary temperature fluctuation amplitude, the vehicle density per unit area, the average temperature value of the cluster, the proportion of high-temperature vehicles and the proportion of abnormal state vehicles are proportionally corrected; The corrected vehicle density per unit area, average cluster temperature, proportion of high-temperature vehicles, proportion of vehicles in abnormal conditions, and charging load ratio of the target area are used together to calculate the risk score of each cluster.

3. The new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling according to claim 1 is characterized in that: The risk level determination module is specifically used to: Obtaining risk score values ​​output by the regional risk modeling module for the same cluster in multiple consecutive time slices, and constructing a score time series in chronological order; Based on the score time series, the incremental value of the risk score between each time slice and its average change rate are calculated to characterize the short-term score growth trend of the cluster; Based on the short-term score growth trend, combined with the current score and static cluster characteristics including vehicle density per unit area and cluster average temperature, the risk score change range within a preset time window is predicted, and the future score prediction value range is output; Perform interval overlap analysis on the future score prediction value interval and the static risk level threshold to identify whether there is an early warning signal of a trend crossing the high risk threshold; When the early warning signal is established, the current score value and the upper limit of the future score prediction value range are combined to construct a trend reinforcement factor and correct the current score value before outputting the final score result; The final scoring result is compared with the risk level determination rule to obtain the risk level currently corresponding to the cluster, and the obtained risk level is used as the output of the risk level determination module.

4. The new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling according to claim 1 is characterized in that: The thermal risk prediction module is specifically used to: Based on the operating status data, a multidimensional time series feature dataset is constructed, which includes the vehicle's geographic location, battery temperature, state of charge, and vehicle operating status labels, and the feature dataset is segmented in chronological order using a sliding window method; A long short-term memory neural network model with a dynamic attention mechanism is used to encode the time series features within each sliding window to obtain a hidden state vector sequence, which is used to capture the temporal patterns and spatial evolution trends of vehicle aggregation behavior. Generate a potential risk score trend for each target area within a preset future time range based on the spatial distribution of the hidden state vector sequence and its relationship with the cluster boundary structure, and output the potential risk score trend in the form of a structured tensor; The output in the form of structured tensors is spatially mapped and converted into a regional-level heat map image format according to the geographic grid encoding rules of the target area. The image is then projected onto the target area map to form a predictive high-risk cluster area layer. The risk score trend of each grid in the high-risk cluster area layer is output to the risk level determination module and the response control scheduling module to complete the scheduling planning of flow control, vehicle guidance and load distribution strategies in the area in advance.

5. The new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling according to claim 2 is characterized in that: The regional risk modeling module is also used to: Based on the spatial boundary morphology of each cluster, the position sequence of the boundary vehicles in a continuous time period is identified, and a boundary trajectory set is constructed, wherein the boundary trajectory set is used to preserve the geographical distribution path of the boundary vehicles over time; Using the boundary trajectory set, the battery temperature of each boundary vehicle in the corresponding time slice is extracted to construct a boundary temperature sequence set with a time index. The boundary temperature sequence set is used to record the dynamic evolution process of the thermal state of the cluster boundary; Based on the boundary temperature sequence set, the temperature variation amplitude, temperature rising rate and continuous temperature difference segment length in each window are extracted in a time sliding window manner to form a local temperature disturbance feature vector sequence; The local temperature disturbance feature vector sequence is subjected to boundary directional aggregation, and the overall boundary temperature disturbance characteristic curve is constructed according to the arrangement order of each boundary segment in the spatial trajectory. The boundary temperature fluctuation amplitude is extracted based on its fluctuation amplitude as output to characterize the thermal stability risk of the cluster structure.

6. The new energy vehicle gathering prevention and control system based on cluster identification and regional risk modeling according to claim 3 is characterized in that: The risk level determination module is further configured to: The future score prediction value interval is decomposed into three indicators: the lower limit value, the upper limit value and the interval span. Based on the short-term score growth trend of the score time series, the distance between the interval span and the current score value is calculated to represent the fluctuation range of the possible future score growth; According to the fluctuation range, all level cutoff values ​​greater than the current score value in the static risk level threshold are retrieved, a set of cutoff values ​​with overlapping values ​​is screened out, and the score difference interval corresponding to the set of cutoff values ​​is output; Compare the score difference interval with the fluctuation range, calculate the coverage ratio of each cutoff value within the fluctuation range, and use the ratio as a trend crossing strength indicator to determine whether the future score interval has a tendency to migrate to a high-risk level; When the trend crossing strength indicator reaches a preset judgment threshold, an early warning signal is generated.

Images